git

1
#define USE_THE_REPOSITORY_VARIABLE
2

3
#include "git-compat-util.h"
4
#include "object-store-ll.h"
5
#include "dir.h"
6
#include "environment.h"
7
#include "gettext.h"
8
#include "hex.h"
9
#include "name-hash.h"
10
#include "sparse-index.h"
11
#include "streaming.h"
12
#include "submodule.h"
13
#include "symlinks.h"
14
#include "progress.h"
15
#include "fsmonitor.h"
16
#include "entry.h"
17
#include "parallel-checkout.h"
18

19
static void create_directories(const char *path, int path_len,
20
			       const struct checkout *state)
21
{
22
	char *buf = xmallocz(path_len);
23
	int len = 0;
24

25
	while (len < path_len) {
26
		do {
27
			buf[len] = path[len];
28
			len++;
29
		} while (len < path_len && path[len] != '/');
30
		if (len >= path_len)
31
			break;
32
		buf[len] = 0;
33

34
		/*
35
		 * For 'checkout-index --prefix=<dir>', <dir> is
36
		 * allowed to be a symlink to an existing directory,
37
		 * and we set 'state->base_dir_len' below, such that
38
		 * we test the path components of the prefix with the
39
		 * stat() function instead of the lstat() function.
40
		 */
41
		if (has_dirs_only_path(buf, len, state->base_dir_len))
42
			continue; /* ok, it is already a directory. */
43

44
		/*
45
		 * If this mkdir() would fail, it could be that there
46
		 * is already a symlink or something else exists
47
		 * there, therefore we then try to unlink it and try
48
		 * one more time to create the directory.
49
		 */
50
		if (mkdir(buf, 0777)) {
51
			if (errno == EEXIST && state->force &&
52
			    !unlink_or_warn(buf) && !mkdir(buf, 0777))
53
				continue;
54
			die_errno("cannot create directory at '%s'", buf);
55
		}
56
	}
57
	free(buf);
58
}
59

60
static void remove_subtree(struct strbuf *path)
61
{
62
	DIR *dir = opendir(path->buf);
63
	struct dirent *de;
64
	int origlen = path->len;
65

66
	if (!dir)
67
		die_errno("cannot opendir '%s'", path->buf);
68
	while ((de = readdir_skip_dot_and_dotdot(dir)) != NULL) {
69
		struct stat st;
70

71
		strbuf_addch(path, '/');
72
		strbuf_addstr(path, de->d_name);
73
		if (lstat(path->buf, &st))
74
			die_errno("cannot lstat '%s'", path->buf);
75
		if (S_ISDIR(st.st_mode))
76
			remove_subtree(path);
77
		else if (unlink(path->buf))
78
			die_errno("cannot unlink '%s'", path->buf);
79
		strbuf_setlen(path, origlen);
80
	}
81
	closedir(dir);
82
	if (rmdir(path->buf))
83
		die_errno("cannot rmdir '%s'", path->buf);
84
}
85

86
static int create_file(const char *path, unsigned int mode)
87
{
88
	mode = (mode & 0100) ? 0777 : 0666;
89
	return open(path, O_WRONLY | O_CREAT | O_EXCL, mode);
90
}
91

92
void *read_blob_entry(const struct cache_entry *ce, size_t *size)
93
{
94
	enum object_type type;
95
	unsigned long ul;
96
	void *blob_data = repo_read_object_file(the_repository, &ce->oid,
97
						&type, &ul);
98

99
	*size = ul;
100
	if (blob_data) {
101
		if (type == OBJ_BLOB)
102
			return blob_data;
103
		free(blob_data);
104
	}
105
	return NULL;
106
}
107

108
static int open_output_fd(char *path, const struct cache_entry *ce, int to_tempfile)
109
{
110
	int symlink = (ce->ce_mode & S_IFMT) != S_IFREG;
111
	if (to_tempfile) {
112
		xsnprintf(path, TEMPORARY_FILENAME_LENGTH, "%s",
113
			  symlink ? ".merge_link_XXXXXX" : ".merge_file_XXXXXX");
114
		return mkstemp(path);
115
	} else {
116
		return create_file(path, !symlink ? ce->ce_mode : 0666);
117
	}
118
}
119

120
int fstat_checkout_output(int fd, const struct checkout *state, struct stat *st)
121
{
122
	/* use fstat() only when path == ce->name */
123
	if (fstat_is_reliable() &&
124
	    state->refresh_cache && !state->base_dir_len) {
125
		return !fstat(fd, st);
126
	}
127
	return 0;
128
}
129

130
static int streaming_write_entry(const struct cache_entry *ce, char *path,
131
				 struct stream_filter *filter,
132
				 const struct checkout *state, int to_tempfile,
133
				 int *fstat_done, struct stat *statbuf)
134
{
135
	int result = 0;
136
	int fd;
137

138
	fd = open_output_fd(path, ce, to_tempfile);
139
	if (fd < 0)
140
		return -1;
141

142
	result |= stream_blob_to_fd(fd, &ce->oid, filter, 1);
143
	*fstat_done = fstat_checkout_output(fd, state, statbuf);
144
	result |= close(fd);
145

146
	if (result)
147
		unlink(path);
148
	return result;
149
}
150

151
void enable_delayed_checkout(struct checkout *state)
152
{
153
	if (!state->delayed_checkout) {
154
		state->delayed_checkout = xmalloc(sizeof(*state->delayed_checkout));
155
		state->delayed_checkout->state = CE_CAN_DELAY;
156
		string_list_init_nodup(&state->delayed_checkout->filters);
157
		string_list_init_nodup(&state->delayed_checkout->paths);
158
	}
159
}
160

161
static int remove_available_paths(struct string_list_item *item, void *cb_data)
162
{
163
	struct string_list *available_paths = cb_data;
164
	struct string_list_item *available;
165

166
	available = string_list_lookup(available_paths, item->string);
167
	if (available)
168
		available->util = item->util;
169
	return !available;
170
}
171

172
static int string_is_not_null(struct string_list_item *item, void *data UNUSED)
173
{
174
	return !!item->string;
175
}
176

177
int finish_delayed_checkout(struct checkout *state, int show_progress)
178
{
179
	int errs = 0;
180
	unsigned processed_paths = 0;
181
	off_t filtered_bytes = 0;
182
	struct string_list_item *filter, *path;
183
	struct progress *progress = NULL;
184
	struct delayed_checkout *dco = state->delayed_checkout;
185

186
	if (!state->delayed_checkout)
187
		return errs;
188

189
	dco->state = CE_RETRY;
190
	if (show_progress)
191
		progress = start_delayed_progress(_("Filtering content"), dco->paths.nr);
192
	while (dco->filters.nr > 0) {
193
		for_each_string_list_item(filter, &dco->filters) {
194
			struct string_list available_paths = STRING_LIST_INIT_DUP;
195

196
			if (!async_query_available_blobs(filter->string, &available_paths)) {
197
				/* Filter reported an error */
198
				errs = 1;
199
				filter->string = NULL;
200
				continue;
201
			}
202
			if (available_paths.nr <= 0) {
203
				/*
204
				 * Filter responded with no entries. That means
205
				 * the filter is done and we can remove the
206
				 * filter from the list (see
207
				 * "string_list_remove_empty_items" call below).
208
				 */
209
				filter->string = NULL;
210
				continue;
211
			}
212

213
			/*
214
			 * In dco->paths we store a list of all delayed paths.
215
			 * The filter just send us a list of available paths.
216
			 * Remove them from the list.
217
			 */
218
			filter_string_list(&dco->paths, 0,
219
				&remove_available_paths, &available_paths);
220

221
			for_each_string_list_item(path, &available_paths) {
222
				struct cache_entry* ce;
223

224
				if (!path->util) {
225
					error("external filter '%s' signaled that '%s' "
226
					      "is now available although it has not been "
227
					      "delayed earlier",
228
					      filter->string, path->string);
229
					errs |= 1;
230

231
					/*
232
					 * Do not ask the filter for available blobs,
233
					 * again, as the filter is likely buggy.
234
					 */
235
					filter->string = NULL;
236
					continue;
237
				}
238
				ce = index_file_exists(state->istate, path->string,
239
						       strlen(path->string), 0);
240
				if (ce) {
241
					display_progress(progress, ++processed_paths);
242
					errs |= checkout_entry(ce, state, NULL, path->util);
243
					filtered_bytes += ce->ce_stat_data.sd_size;
244
					display_throughput(progress, filtered_bytes);
245
				} else
246
					errs = 1;
247
			}
248

249
			string_list_clear(&available_paths, 0);
250
		}
251

252
		filter_string_list(&dco->filters, 0, string_is_not_null, NULL);
253
	}
254
	stop_progress(&progress);
255
	string_list_clear(&dco->filters, 0);
256

257
	/* At this point we should not have any delayed paths anymore. */
258
	errs |= dco->paths.nr;
259
	for_each_string_list_item(path, &dco->paths) {
260
		error("'%s' was not filtered properly", path->string);
261
	}
262
	string_list_clear(&dco->paths, 0);
263

264
	free(dco);
265
	state->delayed_checkout = NULL;
266

267
	return errs;
268
}
269

270
void update_ce_after_write(const struct checkout *state, struct cache_entry *ce,
271
			   struct stat *st)
272
{
273
	if (state->refresh_cache) {
274
		assert(state->istate);
275
		fill_stat_cache_info(state->istate, ce, st);
276
		ce->ce_flags |= CE_UPDATE_IN_BASE;
277
		mark_fsmonitor_invalid(state->istate, ce);
278
		state->istate->cache_changed |= CE_ENTRY_CHANGED;
279
	}
280
}
281

282
/* Note: ca is used (and required) iff the entry refers to a regular file. */
283
static int write_entry(struct cache_entry *ce, char *path, struct conv_attrs *ca,
284
		       const struct checkout *state, int to_tempfile,
285
		       int *nr_checkouts)
286
{
287
	unsigned int ce_mode_s_ifmt = ce->ce_mode & S_IFMT;
288
	struct delayed_checkout *dco = state->delayed_checkout;
289
	int fd, ret, fstat_done = 0;
290
	char *new_blob;
291
	struct strbuf buf = STRBUF_INIT;
292
	size_t size;
293
	ssize_t wrote;
294
	size_t newsize = 0;
295
	struct stat st;
296
	const struct submodule *sub;
297
	struct checkout_metadata meta;
298
	static int scratch_nr_checkouts;
299

300
	clone_checkout_metadata(&meta, &state->meta, &ce->oid);
301

302
	if (ce_mode_s_ifmt == S_IFREG) {
303
		struct stream_filter *filter = get_stream_filter_ca(ca, &ce->oid);
304
		if (filter &&
305
		    !streaming_write_entry(ce, path, filter,
306
					   state, to_tempfile,
307
					   &fstat_done, &st))
308
			goto finish;
309
	}
310

311
	switch (ce_mode_s_ifmt) {
312
	case S_IFLNK:
313
		new_blob = read_blob_entry(ce, &size);
314
		if (!new_blob)
315
			return error("unable to read sha1 file of %s (%s)",
316
				     ce->name, oid_to_hex(&ce->oid));
317

318
		/*
319
		 * We can't make a real symlink; write out a regular file entry
320
		 * with the symlink destination as its contents.
321
		 */
322
		if (!has_symlinks || to_tempfile)
323
			goto write_file_entry;
324

325
		ret = symlink(new_blob, path);
326
		free(new_blob);
327
		if (ret)
328
			return error_errno("unable to create symlink %s", path);
329
		break;
330

331
	case S_IFREG:
332
		/*
333
		 * We do not send the blob in case of a retry, so do not
334
		 * bother reading it at all.
335
		 */
336
		if (dco && dco->state == CE_RETRY) {
337
			new_blob = NULL;
338
			size = 0;
339
		} else {
340
			new_blob = read_blob_entry(ce, &size);
341
			if (!new_blob)
342
				return error("unable to read sha1 file of %s (%s)",
343
					     ce->name, oid_to_hex(&ce->oid));
344
		}
345

346
		/*
347
		 * Convert from git internal format to working tree format
348
		 */
349
		if (dco && dco->state != CE_NO_DELAY) {
350
			ret = async_convert_to_working_tree_ca(ca, ce->name,
351
							       new_blob, size,
352
							       &buf, &meta, dco);
353
			if (ret) {
354
				struct string_list_item *item =
355
					string_list_lookup(&dco->paths, ce->name);
356
				if (item) {
357
					item->util = nr_checkouts ? nr_checkouts
358
							: &scratch_nr_checkouts;
359
					free(new_blob);
360
					goto delayed;
361
				}
362
			}
363
		} else {
364
			ret = convert_to_working_tree_ca(ca, ce->name, new_blob,
365
							 size, &buf, &meta);
366
		}
367

368
		if (ret) {
369
			free(new_blob);
370
			new_blob = strbuf_detach(&buf, &newsize);
371
			size = newsize;
372
		}
373
		/*
374
		 * No "else" here as errors from convert are OK at this
375
		 * point. If the error would have been fatal (e.g.
376
		 * filter is required), then we would have died already.
377
		 */
378

379
	write_file_entry:
380
		fd = open_output_fd(path, ce, to_tempfile);
381
		if (fd < 0) {
382
			free(new_blob);
383
			return error_errno("unable to create file %s", path);
384
		}
385

386
		wrote = write_in_full(fd, new_blob, size);
387
		if (!to_tempfile)
388
			fstat_done = fstat_checkout_output(fd, state, &st);
389
		close(fd);
390
		free(new_blob);
391
		if (wrote < 0)
392
			return error("unable to write file %s", path);
393
		break;
394

395
	case S_IFGITLINK:
396
		if (to_tempfile)
397
			return error("cannot create temporary submodule %s", ce->name);
398
		if (mkdir(path, 0777) < 0)
399
			return error("cannot create submodule directory %s", path);
400
		sub = submodule_from_ce(ce);
401
		if (sub)
402
			return submodule_move_head(ce->name, state->super_prefix,
403
				NULL, oid_to_hex(&ce->oid),
404
				state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
405
		break;
406

407
	default:
408
		return error("unknown file mode for %s in index", ce->name);
409
	}
410

411
finish:
412
	if (state->refresh_cache) {
413
		if (!fstat_done && lstat(ce->name, &st) < 0)
414
			return error_errno("unable to stat just-written file %s",
415
					   ce->name);
416
		update_ce_after_write(state, ce , &st);
417
	}
418
	if (nr_checkouts)
419
		(*nr_checkouts)++;
420
delayed:
421
	return 0;
422
}
423

424
/*
425
 * This is like 'lstat()', except it refuses to follow symlinks
426
 * in the path, after skipping "skiplen".
427
 */
428
static int check_path(const char *path, int len, struct stat *st, int skiplen)
429
{
430
	const char *slash = path + len;
431

432
	while (path < slash && *slash != '/')
433
		slash--;
434
	if (!has_dirs_only_path(path, slash - path, skiplen)) {
435
		errno = ENOENT;
436
		return -1;
437
	}
438
	return lstat(path, st);
439
}
440

441
static void mark_colliding_entries(const struct checkout *state,
442
				   struct cache_entry *ce, struct stat *st)
443
{
444
	int i, trust_ino = check_stat;
445

446
#if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__)
447
	trust_ino = 0;
448
#endif
449

450
	ce->ce_flags |= CE_MATCHED;
451

452
	/* TODO: audit for interaction with sparse-index. */
453
	ensure_full_index(state->istate);
454
	for (i = 0; i < state->istate->cache_nr; i++) {
455
		struct cache_entry *dup = state->istate->cache[i];
456

457
		if (dup == ce) {
458
			/*
459
			 * Parallel checkout doesn't create the files in index
460
			 * order. So the other side of the collision may appear
461
			 * after the given cache_entry in the array.
462
			 */
463
			if (parallel_checkout_status() == PC_RUNNING)
464
				continue;
465
			else
466
				break;
467
		}
468

469
		if (dup->ce_flags & (CE_MATCHED | CE_VALID | CE_SKIP_WORKTREE))
470
			continue;
471

472
		if ((trust_ino && !match_stat_data(&dup->ce_stat_data, st)) ||
473
		    paths_collide(ce->name, dup->name)) {
474
			dup->ce_flags |= CE_MATCHED;
475
			break;
476
		}
477
	}
478
}
479

480
int checkout_entry_ca(struct cache_entry *ce, struct conv_attrs *ca,
481
		      const struct checkout *state, char *topath,
482
		      int *nr_checkouts)
483
{
484
	static struct strbuf path = STRBUF_INIT;
485
	struct stat st;
486
	struct conv_attrs ca_buf;
487

488
	if (ce->ce_flags & CE_WT_REMOVE) {
489
		if (topath)
490
			/*
491
			 * No content and thus no path to create, so we have
492
			 * no pathname to return.
493
			 */
494
			BUG("Can't remove entry to a path");
495
		unlink_entry(ce, state->super_prefix);
496
		return 0;
497
	}
498

499
	if (topath) {
500
		if (S_ISREG(ce->ce_mode) && !ca) {
501
			convert_attrs(state->istate, &ca_buf, ce->name);
502
			ca = &ca_buf;
503
		}
504
		return write_entry(ce, topath, ca, state, 1, nr_checkouts);
505
	}
506

507
	strbuf_reset(&path);
508
	strbuf_add(&path, state->base_dir, state->base_dir_len);
509
	strbuf_add(&path, ce->name, ce_namelen(ce));
510

511
	if (!check_path(path.buf, path.len, &st, state->base_dir_len)) {
512
		const struct submodule *sub;
513
		unsigned changed = ie_match_stat(state->istate, ce, &st,
514
						 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
515
		/*
516
		 * Needs to be checked before !changed returns early,
517
		 * as the possibly empty directory was not changed
518
		 */
519
		sub = submodule_from_ce(ce);
520
		if (sub) {
521
			int err;
522
			if (!is_submodule_populated_gently(ce->name, &err)) {
523
				struct stat sb;
524
				if (lstat(ce->name, &sb))
525
					die(_("could not stat file '%s'"), ce->name);
526
				if (!(st.st_mode & S_IFDIR))
527
					unlink_or_warn(ce->name);
528

529
				return submodule_move_head(ce->name, state->super_prefix,
530
					NULL, oid_to_hex(&ce->oid), 0);
531
			} else
532
				return submodule_move_head(ce->name, state->super_prefix,
533
					"HEAD", oid_to_hex(&ce->oid),
534
					state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
535
		}
536

537
		if (!changed)
538
			return 0;
539
		if (!state->force) {
540
			if (!state->quiet)
541
				fprintf(stderr,
542
					"%s already exists, no checkout\n",
543
					path.buf);
544
			return -1;
545
		}
546

547
		if (state->clone)
548
			mark_colliding_entries(state, ce, &st);
549

550
		/*
551
		 * We unlink the old file, to get the new one with the
552
		 * right permissions (including umask, which is nasty
553
		 * to emulate by hand - much easier to let the system
554
		 * just do the right thing)
555
		 */
556
		if (S_ISDIR(st.st_mode)) {
557
			/* If it is a gitlink, leave it alone! */
558
			if (S_ISGITLINK(ce->ce_mode))
559
				return 0;
560
			/*
561
			 * We must avoid replacing submodules' leading
562
			 * directories with symbolic links, lest recursive
563
			 * clones can write into arbitrary locations.
564
			 *
565
			 * Technically, this logic is not limited
566
			 * to recursive clones, or for that matter to
567
			 * submodules' paths colliding with symbolic links'
568
			 * paths. Yet it strikes a balance in favor of
569
			 * simplicity, and if paths are colliding, we might
570
			 * just as well keep the directories during a clone.
571
			 */
572
			if (state->clone && S_ISLNK(ce->ce_mode))
573
				return 0;
574
			remove_subtree(&path);
575
		} else if (unlink(path.buf))
576
			return error_errno("unable to unlink old '%s'", path.buf);
577
	} else if (state->not_new)
578
		return 0;
579

580
	create_directories(path.buf, path.len, state);
581

582
	if (S_ISREG(ce->ce_mode) && !ca) {
583
		convert_attrs(state->istate, &ca_buf, ce->name);
584
		ca = &ca_buf;
585
	}
586

587
	if (!enqueue_checkout(ce, ca, nr_checkouts))
588
		return 0;
589

590
	return write_entry(ce, path.buf, ca, state, 0, nr_checkouts);
591
}
592

593
void unlink_entry(const struct cache_entry *ce, const char *super_prefix)
594
{
595
	const struct submodule *sub = submodule_from_ce(ce);
596
	if (sub) {
597
		/* state.force is set at the caller. */
598
		submodule_move_head(ce->name, super_prefix, "HEAD", NULL,
599
				    SUBMODULE_MOVE_HEAD_FORCE);
600
	}
601
	if (check_leading_path(ce->name, ce_namelen(ce), 1) >= 0)
602
		return;
603
	if (remove_or_warn(ce->ce_mode, ce->name))
604
		return;
605
	schedule_dir_for_removal(ce->name, ce_namelen(ce));
606
}
607

608
int remove_or_warn(unsigned int mode, const char *file)
609
{
610
	return S_ISGITLINK(mode) ? rmdir_or_warn(file) : unlink_or_warn(file);
611
}
612